Quick answer
Starrd's Deep Down In My Heart template recreates the viral crying-interview meme from two photos. Photo 1 is the crier, who sits still, presses a hand over their mouth, then breaks down wailing with tears pouring and their chin in their hand. Photo 2 is the confessor, who says "Because I know somewhere deep down in my heart… I still love you" and then laughs through tears. The full 30-second clip is rebuilt shot for shot, all eight cuts on the original's timing, with the original audio and real lip sync, in the place from the crier's photo. Works with people and pets. Vertical 9:16, Seedance 2.5, 250 credits (about $12.50).
Two photos in, and out comes you and a friend recreating the "deep down in my heart… I still love you" interview: one of you confesses, the other holds it in and then breaks down in the loudest, ugliest cry on the internet. All 30 seconds, lip-synced to the original audio. Here it is with two cats:
Deep Down In My Heart
Two photos. One of you says 'I still love you', the other breaks down in the ugliest cry on the internet. The full 30-second meme, lip-synced.
What You Get
- The whole scene. All 30 seconds and all eight cuts, at the same moments as the original.
- Two roles that never swap. Photo 1 is the crier, photo 2 is the confessor.
- Real lip sync. The confessor mouths both lines, the crier stays silent until the breakdown, then wails in time with every cry.
- A real ugly cry. The hand over the mouth, the face crumpling, tears streaming, the wide-open wail, the chin in the hand. The confessor's eyes well up too.
- Your own place. The scene happens in the background of the crier's photo.
- People or pets. Cats and dogs cry with their own faces and tear-soaked fur.
- 30 seconds, vertical 9:16, ready for TikTok, Reels and Shorts.
What the Meme Actually Is
It's an old documentary interview that keeps coming back. Two men sit side by side on a brown leather couch in a living room. One, in glasses and a striped track jacket, chokes out "Because I know somewhere deep down in my heart…", and after a long pause, "I still love you." The other has been sitting with a hand over his mouth the whole time. Then he breaks: a long, open-mouthed wail that goes on for ten seconds while the first one half laughs, half cries.
In October 2026 @nozlive999 reposted it as "No way, I found the original video", and it passed 100,000 plays and 7,000 shares within days.
Why it suits AI: the whole joke is two faces and one sound. Two photos let you cast your friend, your partner or your cat as the crier without anyone having to actually cry on camera.
The Scene, Shot by Shot
| Time | Shot | What happens |
|---|---|---|
| 0:00 | Two-shot | The crier faces the camera, guarded; the confessor is the back of a head: "Because I know…" |
| 1.6s | Confessor close-up | "…somewhere deep down in my heart," then a teary half-smile |
| 6.8s | Crier close-up | Silent, a hand pressed over the mouth, wet eyes |
| 9.1s | Confessor close-up | "I still love you." |
| 13.4s | Crier close-up | The face crumples, then the breakdown: the loudest wail at 19s, the chin in a hand from 21s |
| 25.5s | Crier, tighter | Still sobbing |
| 26.6s | Confessor close-up | Laughing through tears |
| 29.4s | Crier | Sobbing into one hand |
Who plays who: photo 1 cries, photo 2 talks. Put the friend who'd be funniest falling apart first.
Your Look, Your Place
- Your own face, hair and clothes. Nobody gets the original's glasses or track jacket.
- Your own glasses. If your photo has sunglasses, you keep them on; if it doesn't, you don't get any.
- Your own background. The sofa scene is built in the place from the crier's photo, in the same light.
Use clear, well-lit face photos. Almost every shot is a close-up, so faces carry the video. Sunglasses stay on if your photo has them, which hides some of the tears, so a photo without them gives the biggest ugly cry.
The Fastest Way
Deep Down In My Heart
Two photos, the full 30-second crying meme with the original audio, lip-synced.
Two photos, 250 credits (about $12.50), no prompt. It renders on Seedance 2.5 at 480p, vertical 9:16.
Build It Yourself
1. Reframe the clip to vertical. The original is square. Crop each frame to 9:16 around the face, following it as the person rocks and leans, and never smoothing the crop across a cut.
2. Scrub the burned-in subtitles and app icons. Their letters show up as raised shapes in a depth map. Find the thin bright caption strokes and the on-screen icons in each source frame and inpaint them before you make the depth map.
3. Turn it into a depth map, one shot at a time. Run it through a depth-estimation model (Depth Anything V2 works) and set the contrast per shot, then put the original audio inside the depth video.
4. Make one storyboard sheet with a panel per beat. The two-shot, the confessor choked up, the crier's hand over the mouth, the confessor laughing through tears, the wail, and the sob into one hand, all from the two photos.
A close-up of the person from reference image 1 in three-quarter view, the face turned toward the right edge of the frame: an ugly cry, the whole face scrunched, the brows pulled up hard in the middle, the eyes squeezed shut, the mouth stretched wide open in a loud wail, streams of tears down both cheeks, a runny nose, the skin flushed and blotchy. No text.
5. Drive it with the depth map and the sound. Feed the sheet, both photos and the depth video to a model that copies motion from a reference video (we recommend Seedance 2.5). Then write three things it can't get from a depth map:
- A timed shot list: who is on screen and what they do for each of the eight cuts.
- The speaking windows: when the confessor's mouth moves, and that the crier stays silent until the breakdown.
- The tears: say exactly how ugly the cry is, or the model gives you a polite, dry-eyed sniffle.
Lay the original audio back on at the end.
Common Mistakes
Driving with the raw clip. The original faces and outfits come straight through. Use the depth map.
Leaving the subtitles in. Faint letter shapes in the depth map are enough for the model to invent text.
Not pinning the short shots. The crier's hand-over-the-mouth shot is only 2.3 seconds. If you don't say where it cuts, the model holds it a second too long and "I still love you" starts over the wrong face.
Pretty crying. Without tears, a runny nose and a scrunched face written into the prompt, the breakdown reads as mild sadness.
Animal hands. A depth map of a human hand can turn a pet's paw into a hand. Tell the model the animal's front legs end in real paws.
FAQ
How many photos? Exactly two: the crier and the confessor.
Does it work with pets? Yes, cats and dogs included.
How long is it? 30 seconds, vertical 9:16.
What does it cost? 250 credits, about $12.50.
Related Reading
- How to Make the 'Like 700 Mile Per Hour' AI Interview Meme: another two-person interview meme, lip-synced line for line
- How to Make the 'Hit the Road Dude' AI Video: another viral clip, lip-synced line for line
- How to Make the 'How I Become a Millionaire' AI Video: the ChatGPT voice-memo skit, starring you
- Viral AI Video Trends 2026: every trend we've turned into a template this year
